Research On Fault Diagnosability Evaluation And Method Of Fault Diagnosis For Nonlinear System

The research shows that the low diagnosability of the fault is one of the main causes of the system fault,that is,the measurement information is not enough to support the rapid and reliable detection of the system fault.For the study of fault diagnosability,most of the traditional methods focus on the qualitative research after the occurrence of fault,which can neither quantify the true level of fault diagnosability nor improve it.Therefore,it is urgent to explore how to determine the fault diagnostic performance of the system in the design stage(before the fault occurs),how to make the system have higher fault diagnosis capability through the optimized configuration of sensors in the design stage,and how to select appropriate methods for effective fault diagnosis of the system.In view of the above problems,the following work has been carried out in this dissertation.1)Research on quantitative evaluation method for fault diagnosability of nonlinear systemsIn view of the nonlinear system fault diagnosis can be lack of quantitative evaluation of problem,based on the K-L divergence algorithm,fully considering the disturbance and noise uncertainty factors that affect the accuracy of the evaluation results,with sparse kernel density estimation and monte carlo method as beneficial supplement,through to the single fault or under different fault condition probability density function of the similarity and difference degree calculation,given for nonlinear system fault accurately solving method and steps of the quantitative evaluation,and the fault diagnostic evaluation index should be measured noise feasible domain analysis,further fault diagnosis design for the system and laid the foundation of fault diagnosis.2)Research on the design method of fault diagnosability for nonlinear systemIn order to augment the measurement information of the system and improve the fault diagnostic evaluation index of the system,the greedy algorithm is used comprehensively to optimize the design of the measuring point sensor,and the soft sensor is constructed by means of KPLS method to realize the design of measuring point sensor with soft instead of hard.The system can meet the basic requirements of fault diagnosis through the optimal configuration of detection point information,so the fault diagnosis performance of the system is incorporated into the system design at the beginning,which provides an effective way to improve the safety level of the system.3)Research on optimal sensor placement method based on quantitative evaluation of fault diagnosabilityIn order to optimize the configuration of measuring point sensor,a sensor optimization configuration model is designed,which focuses on quantitative evaluation and takes into account optimization objectives such as reliability and economy.The nonlinear dynamic programming and the improved NSGA-II algorithm are used to make the optimal configuration process of sensors more efficient from a quantitative perspective,providing a referential method and reference basis for high quality and low cost system operation in reality.4)Research on fault diagnosis of nonlinear systems based on model methodA fault detection and separation method based on particle filter is designed on the basis of satisfying the diagnostic evaluation index of fault.At the same time,an adaptive threshold design method based on the statistical characteristic analysis of residual error is proposed,which can effectively reduce the missing report rate and false report rate of fault diagnosis without depending on the system mechanism and mathematical model.5)Research on non-stationary process fault detection method based on data drivenThe classification number of the system residual probability density function is determined by the mode adaption method,and the k-mean method is used to cluster the residual data.Then,by using K-L divergence method,the distance difference between the real-time data of the system and the offline data residual probability density function is calculated online,and the fault detection is carried out through the classification evaluation of non-stationary residual process with multi-mode operation system.Compared with linear systems,non-linear systems,due to the complexity of their construction and the presence of uncertainties such as noise and external disturbance,are always difficult problems in the research on fault diagnosability.And practical engineering system are mostly exists a certain degree of nonlinear characteristics,therefore,in view of the nonlinear system,this dissertation explore the quantitative evaluation of the fault diagnosis can be an effective method to analyze the key factors to fault diagnosis can be,the study has a practical fault diagnosis can be improve design scheme,and be included in the system,the design of the system,on the basis of diagnostic as the goal,in order to improve the system can optimize the sensor configuration,and system fault diagnosis scheme is given,to improve the quality system,to promote modern engineering towards a stable,efficient and economical development is of great significance.